Scene Evaluation/Collecting Nodes time VS Standalone

[justix]

Once again I'm here writing concerning my problems ( sorry I must say ) about the actual version that is keeping me not to operate safely knowing all the facts about my own GPU limits 2 x 1070 but basically only 8GB and handling large Polygons + texures that may apply ( but I understand the CPU out of core can help ..not on the geometry? )

I had all sort of hanging and crashes and bl..dy autosave in the middle of it and generally speaking..instability with a scene that a client of mine passed me over ( leave the fact that I had to work like a madman just to re-texture it all ( corona ) but also to Crunch/compress most of the HIgh Poly models in the scene to cut a decent camera view keeping the poly count to max 5-6 Millions Polygons in order to help me work .. touching 10 Millions is just making the whole system unstable with hangs and huge lags...but not on the standalone-

I stick to my best Polycount to 7 Millions but still the standalone is WAY BETTER than the Plugin..why is that? I mean a test which I can share ( a free model 100K ) copied up to 7Millions polygons takes 46 seconds ( if it does not hang ) with the Counting of the nodes that get stucked most of the times

7.197.408 Triangles on the Max plugin = 46 Seconds when lucky as 3Ds is terrible RAM eater and at least with this vers. ( and the last before this the 4.05-6.20 ) a pain

7.197.408 Triangles on the 4.05 Standalone = 8 Seconds an no major hang

Tested many times, resetting the PC freeing up RAM ect... same results

PLUS the same personal scene ( about 4 Mil polygons ) is working fine exporting it to the Standalone but no chance on the 3dmax side..it basically hang upon ending of Collecting nodes and start dumping on the Hard Disk like madness..I can only reset.. annoying

[paride4331]

Hi justix,
Autosave doesn't depend on Octane, you can disable it: 3dsmax > File > preferences > File > uncheck "enable".
Cpu memory feature (Octane V2018.x and V2019.x) lets you use more textures and geometry (not geometry only) than would fit in VRAM by keeping them in RAM instead; this feature increases and any RAM occupied with cpu memory data will not be available to other applications. This means would be recommendable have sufficient RAM, personally I recommend minimum 64 GB.
You can optimize DXTC textures in RGB texture node.
Ram is the reason for the tests you did, 3dsMax needs RAM then Octane also needs RAM. So you have less RAM if you only use Octane Standalone instead. Furthermore OctaneRender Standalone does not use opengl so it also needs less resources.
Regards
Paride

[justix]

Hi Paride.
I know I did open a similar thread a long while ago but I'm on V 4.05 - 6.21 and can't test 2018-2019 as on a STAND SUB T3 and so I'm stuck I presume, I'm fully aware of all RAM problems ect and my issues are concerning also the difference between plugins and..plugins, I get that OTOY is relying on a Out of Core solution for the geometry + texture VS Vram instead of a sort of optimization like I see elsewhere ( Centileo for example is claiming 750M unique polygons for 10GB GPU memory and tested with 15 Mil Triangles on my 8GB 1070 and did not even hang a bit eating it up in 20 sec.. on my max 2016.. I mean not standalone ).

But at least on 4.05-6.21 the CPU Out of core set at 8GB makes no difference at all..it takes more than 60 seconds to load geometry ( 11Mil Triangles ) an then freezes up while I guess the CPU out of core option should take care of the extra geometry?

Please try to read my passion for it...because we all love to have a competitive system somehow... I take the chance to leave a little test here so that anyone with 8gb card can test it and report if they would... like 7 Mil Triangles or copying the msh up to 10-15 Mil and report back? Is there a Table that we can use to understand how Octane is handling UP TO ( only Vram ) in terms of Polycount? would benefit I guess....

On the positive Notes the SA 4.05 is handling that test well so far.. But again I think ( MHO ) that the plugin need to be optimize to say the least OR I don't undertand why 3Dsmax should eat my Vram instead of the CPU?

PLUS differences between the 4.05 SA with a 16 Mil scene and the 2019 1 RC5 SA the 2019 eems to use less Vram memory compared to 4.05

[paride4331]

Hi Justix,
try setting all your scene to movable proxy.
You could test Octane for 3dsMax V2019.x and V2018,x demo versions available from the website demo page [here](https://home.otoy.com/render/octane-render/demo/).
Regards
Paride

[coilbook]

Hi Justix,
try setting all your scene to movable proxy.
You could test Octane for 3dsMax V2019.x and V2018,x demo versions available from the website demo page [here](https://home.otoy.com/render/octane-render/demo/).
Regards
Paride

Hi Paride,
Will they fix this bug or we have to set everything to movable proxys all the time?

Can otoy develop a system where it has no static or movable proxies? So if I animate something or if something linked moves it knows automatically. Also please add an option enable motion blur for all movable proxys so we don't have to manually do it for each object.
Thank you

[paride4331]

Hi coilbook,
yes, it makes sense, I will ask to developers a movable proxy global setting.
Since the tests with 3.08 in general, I can say in my experience, evaluation time is shorter using all movable proxy; presumably, I'm not a developer, it depends on GPU transfer times. Octane takes long to create a large package to send to GPU instead of small packages and small packages are faster to send to GPU. Maybe render time, using non movable proxy, could be shorter, in simple animated or static render scenes.
Regards
Paride

[tcheng00]

Hi everyone. Thanks for this thread, but it sounds like I should always have everything set to movable proxies for performance reasons.

[paride4331]

Hi everyone. Thanks for this thread, but it sounds like I should always have everything set to movable proxies for performance reasons.

viewtopic.php?f=27&t=74536&start=10#p381421